Denoising autoencoders for fast real-Time traffic estimation on urban road networks

We propose a new method for traffic state estimation applicable to large urban road networks where a significant amount of the real-Time and historical data is missing. Our proposed approach involves estimating the missing historical data through low-rank matrix completion, coupled with an online estimation approach for estimating the missing real-Time data. In contrast to the traditional approach, the proposed method does not require re-calibration every time new streaming data becomes available. Empirical results from two metropolitan cities show that the proposed two-step approach provides comparable accuracy to a state of the art benchmark method while achieving two orders of magnitude improvement in computational speed.